Feasibility Study and R&D
Feasibility study and R&D activities: your AI from scratch
A structured programme to find out whether an AI idea really works on your data, before committing production budget. Fixed price, concrete deliverable, code release option exercisable only after seeing the results.
Who it is for
Companies that want to validate an AI idea seriously before investing production budget. R&D teams, innovation managers, IT directors with a concrete problem: sales forecasting, document analysis, computer vision, NLP on proprietary corpora. For those who need to find out whether AI can really solve it, on which data, with which limits, without signing binding development contracts.
Three phases plus one. Total transparency on the output.
An honest method: first we explore, then we decide, and only if the results convince do we release the code.
Phase 1: Preliminary scoping sessions (FREE)
We identify the primary need, the problem statement, the available data and the compliance constraints. From here we produce a formal quote with perimeter and timing.
Phase 2: R&D experimentation
Real technical work on models, benchmarks, configurations. They can be classical statistical models, specialised LLMs, hybrid systems combining interpretability and predictive power. The plan can evolve: iterative R&D is part of the method.
Phase 3: Assessment Report
Summary document delivered within 30 days of the end of the experiments. Contains the results obtained, the benchmarks performed, the confidence metrics and an objective, honest assessment of the current ability of AI models to solve the identified task. Even a 'doesn't work well enough' is a valuable output: it saves you from spending the production budget.
Phase 4 (OPTIONAL): Decision and Code Release
Based on the report you decide whether the R&D has confirmed feasibility or not. No pressure to continue, no hidden strings. If the results are in line with your expectations, you exercise the option and receive scripts, source code of the tested algorithms and configuration metrics. If you don't exercise it, the engagement closes with the report and nothing more is owed.
Why this model is different
Obligation of means, not of result
R&D, by its nature, explores. We guarantee method, rigour and professional diligence, not the existence of a perfect solution that may not exist yet. It's the honest way to do AI research.
Fixed price, no surprises
Three payment milestones, the last of which is optional and depends on your evaluation of the report.
The code is yours, but it's your choice
The final 20% is the code release option. You exercise it only if you've read the report and you're convinced. You avoid paying for code you'll never use.
Native AI Act compliance
In compliance with Regulation (EU) 2024/1689, we preliminarily qualify whether the use case falls into 'High Risk' sectors. The compliance procedures for any production deployment are governed by a separate agreement.
Frequently asked questions
We develop different types of algorithms; over the years we've built: sales forecasting, large-scale document analysis, object recognition, entity classification, recommendation systems and ranking on enterprise data.
Results obtained, benchmarks with documented methodology, confidence metrics, limits encountered and an objective assessment of the current ability of AI models to solve the task. It's a document you can bring to the board to decide whether to proceed.
It's a legitimate outcome of R&D and it's exactly why this model exists. You pay only the deposit and report delivery, you don't exercise the code option, and you avoid committing budget to a production that wouldn't have worked. The report has value in itself.
If you exercise the release option, yes: you receive scripts and source code with the configuration metrics. For critical production deployment (SLA, maintenance, monitoring) a separate agreement is in place: the Feasibility Study does not cover those aspects.
The datasets and materials you provide for testing remain your exclusive property. They are not used to train third-party models without your consent.
Until the release option is exercised, the code developed ad hoc remains AIDAPT's property. Methodological know-how, base frameworks (including Caity), pre-existing libraries and orchestration techniques always remain AIDAPT's.
The experimentation phase depends on the use case and data availability (typically no more than 90 days). The Assessment Report is drafted within 30 days of the end of the experiments. Exact timing is set in the quote after the preliminary sessions.
Got an AI idea to validate?
We start with a preliminary scoping session, no commitment, to see if it makes sense to work together. If the conditions are right, we formalise a fixed-price quote with perimeter, deliverables and defined timelines. No integration, deployment or infrastructure: priority on the algorithm.
